About me

Hi, I am Wei Zhen. I have worked in AI research and engineering since 2018.

My broader interest is in how models learn from complex real-world signals and processes such as speech & language, images and movement. My work has spanned both generation and perception: neural speech synthesis, LLM behavior shaping, visual information extraction and trajectory simulation. Across these domains, I am especially interested in how models learn from imperfect supervision and how meaningful controls of model can be exposed for practical usage.

I started my career at Lyrebird in Montreal (later acquired by Descript), where I worked on neural speech synthesis, voice cloning, and AI speech editing. I moved to Singapore in 2021 and worked on eKYC computer vision problems at Advance AI. I led AI development for document verification technology used in production across Southeast Asia. I later joined Motion G, where I built LLM agents and systems for industrial motion-control engineering workflows and business processes.

I am currently working independently on controllable generative modeling for sequential data. I am exploring applied research opportunities on real-world signals across modalities, especially where practical constraints shape the modeling problems.

You can find more of my work on GitHub, Hugging Face and Google Scholar. You can contact me about my technical work via X and LinkedIn.